Research On Visual Stitching Methold Based On Improved ORB Algorithm

Because the process of image and video stitching are easily disturbed by the external environment,many problems need to be overcome in the image and video splicing.Such as,the real-time stitching,the scene diversity and video jitter.Image registration,image fusion and video stitching method was studied in this thesis.The main contents are as follows:(1)In view of the original ORB(Oriented fast and Rotated Brief)image registration method has good timeliness,but the registration accuracy is poor.Based on the SURF(Speeded Up Robust Features)method with high registration quality and ORB image registration method,an improved ORB image registration method was proposed in this thesis.The ideal of this algorithm is as follows:Firstly,the Hessian matrix method was used to detect feature points.Secondly,the feature descriptors were generated by the BRIEF(Binary Robust Independent Elementary Features)method.Finally,the Hamming distance method was used to rough match.The experimental results show that the proposed method not only overcomes the defect that the original ORB method does not have scale invariance,but also keeps the characteristics of fast registration speed of the original ORB method,and high quality of registration of the SURF method.The registration speed of this algorithm is better than the original SURF method,and the registration accuracy is higher than the original ORB method,which can meet the requirements of video splicing.(2)Because existing fusion methods are severe demanding for the scene,such as,the fusion method of the maximum value requires high quality of the captured image,the fade-in-and-fade-out image fusion algorithm method cannot well handle the scene of the moving object at the splicing place.Therefore,a kind of image fusion method was proposed based on threshold combining with maximum and gradual fading fusion methods.The ideal of this algorithm is as follows:Firstly,the non-fusion image was divided into overlapping and non-overlapping regions.Secondly,the average weighted and the smoothing algorithm with threshold was used to obtain the fusion pixel value of the non-overlapping,the maximum value and the smoothing algorithm with threshold was used to obtain the fusion pixel value of overlapped regions.Finally,the fusion image was generated.The experimental results show that this method can not only eliminate ghosting at the splice,but also obtain the high quality fusion image.(3)Based on the above research,this thesis used the key technology of template transformation matrix to develop a video stitching system.The real-time splicing of three different scenes were completed based on video stitching system.The experimental results show that the video stitching system has certain stability and reliability.